blob: 7394e9e17534ecb03573e28b2288d06cccf24a32 [file] [log] [blame]
Jens Axboe5274f052006-03-30 15:15:30 +02001/*
2 * "splice": joining two ropes together by interweaving their strands.
3 *
4 * This is the "extended pipe" functionality, where a pipe is used as
5 * an arbitrary in-memory buffer. Think of a pipe as a small kernel
6 * buffer that you can use to transfer data from one end to the other.
7 *
8 * The traditional unix read/write is extended with a "splice()" operation
9 * that transfers data buffers to or from a pipe buffer.
10 *
11 * Named by Larry McVoy, original implementation from Linus, extended by
Jens Axboec2058e02006-04-11 13:56:34 +020012 * Jens to support splicing to files, network, direct splicing, etc and
13 * fixing lots of bugs.
Jens Axboe5274f052006-03-30 15:15:30 +020014 *
Jens Axboe0fe23472006-09-04 15:41:16 +020015 * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk>
Jens Axboec2058e02006-04-11 13:56:34 +020016 * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org>
17 * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu>
Jens Axboe5274f052006-03-30 15:15:30 +020018 *
19 */
20#include <linux/fs.h>
21#include <linux/file.h>
22#include <linux/pagemap.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020023#include <linux/splice.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080024#include <linux/memcontrol.h>
Jens Axboe5274f052006-03-30 15:15:30 +020025#include <linux/mm_inline.h>
Jens Axboe5abc97a2006-03-30 15:16:46 +020026#include <linux/swap.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020027#include <linux/writeback.h>
28#include <linux/buffer_head.h>
Jeff Garzika0f06782006-03-30 23:06:13 -050029#include <linux/module.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020030#include <linux/syscalls.h>
Jens Axboe912d35f2006-04-26 10:59:21 +020031#include <linux/uio.h>
James Morris29ce2052007-07-13 11:44:32 +020032#include <linux/security.h>
Jens Axboe5274f052006-03-30 15:15:30 +020033
Jens Axboe83f91352006-04-02 23:05:09 +020034/*
35 * Attempt to steal a page from a pipe buffer. This should perhaps go into
36 * a vm helper function, it's already simplified quite a bit by the
37 * addition of remove_mapping(). If success is returned, the caller may
38 * attempt to reuse this page for another destination.
39 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020040static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020041 struct pipe_buffer *buf)
42{
43 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020044 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020045
Jens Axboe9e0267c2006-04-19 15:57:31 +020046 lock_page(page);
47
Jens Axboe9e94cd42006-06-20 15:01:12 +020048 mapping = page_mapping(page);
49 if (mapping) {
50 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020051
Jens Axboe9e94cd42006-06-20 15:01:12 +020052 /*
53 * At least for ext2 with nobh option, we need to wait on
54 * writeback completing on this page, since we'll remove it
55 * from the pagecache. Otherwise truncate wont wait on the
56 * page, allowing the disk blocks to be reused by someone else
57 * before we actually wrote our data to them. fs corruption
58 * ensues.
59 */
60 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020061
David Howells266cf652009-04-03 16:42:36 +010062 if (page_has_private(page) &&
63 !try_to_release_page(page, GFP_KERNEL))
Jens Axboeca39d652008-05-20 21:27:41 +020064 goto out_unlock;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020065
Jens Axboe9e94cd42006-06-20 15:01:12 +020066 /*
67 * If we succeeded in removing the mapping, set LRU flag
68 * and return good.
69 */
70 if (remove_mapping(mapping, page)) {
71 buf->flags |= PIPE_BUF_FLAG_LRU;
72 return 0;
73 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020074 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020075
Jens Axboe9e94cd42006-06-20 15:01:12 +020076 /*
77 * Raced with truncate or failed to remove page from current
78 * address space, unlock and return failure.
79 */
Jens Axboeca39d652008-05-20 21:27:41 +020080out_unlock:
Jens Axboe9e94cd42006-06-20 15:01:12 +020081 unlock_page(page);
82 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020083}
84
Jens Axboe76ad4d12006-05-03 10:41:33 +020085static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020086 struct pipe_buffer *buf)
87{
88 page_cache_release(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020089 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020090}
91
Jens Axboe08457182007-06-12 20:51:32 +020092/*
93 * Check whether the contents of buf is OK to access. Since the content
94 * is a page cache page, IO may be in flight.
95 */
Jens Axboecac36bb2007-06-14 13:10:48 +020096static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
97 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +020098{
99 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200100 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200101
102 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200103 lock_page(page);
104
105 /*
106 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200107 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200108 */
109 if (!page->mapping) {
110 err = -ENODATA;
111 goto error;
112 }
113
114 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200115 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200116 */
117 if (!PageUptodate(page)) {
118 err = -EIO;
119 goto error;
120 }
121
122 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200123 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200124 */
Jens Axboe5274f052006-03-30 15:15:30 +0200125 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200126 }
127
Jens Axboef84d7512006-05-01 19:59:03 +0200128 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200129error:
130 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200131 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200132}
133
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800134static const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200135 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200136 .map = generic_pipe_buf_map,
137 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb2007-06-14 13:10:48 +0200138 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200139 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200140 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200141 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200142};
143
Jens Axboe912d35f2006-04-26 10:59:21 +0200144static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
145 struct pipe_buffer *buf)
146{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200147 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
148 return 1;
149
Jens Axboe14328732006-05-03 10:35:26 +0200150 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200151 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200152}
153
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800154static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200155 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200156 .map = generic_pipe_buf_map,
157 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb2007-06-14 13:10:48 +0200158 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200159 .release = page_cache_pipe_buf_release,
160 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200161 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200162};
163
Jens Axboe932cc6d2007-06-21 13:10:21 +0200164/**
165 * splice_to_pipe - fill passed data into a pipe
166 * @pipe: pipe to fill
167 * @spd: data to fill
168 *
169 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200170 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200171 * the struct pipe_buf_operations associated with these pages. This
172 * function will link that data to the pipe.
173 *
Jens Axboe83f91352006-04-02 23:05:09 +0200174 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200175ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
176 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200177{
Jens Axboe00de00b2007-06-15 13:14:22 +0200178 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200179 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200180
181 ret = 0;
182 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200183 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200184
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200185 pipe_lock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200186
Jens Axboe5274f052006-03-30 15:15:30 +0200187 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200188 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200189 send_sig(SIGPIPE, current, 0);
190 if (!ret)
191 ret = -EPIPE;
192 break;
193 }
194
Jens Axboe6f767b02006-04-11 13:53:56 +0200195 if (pipe->nrbufs < PIPE_BUFFERS) {
196 int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200197 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200198
Jens Axboe912d35f2006-04-26 10:59:21 +0200199 buf->page = spd->pages[page_nr];
200 buf->offset = spd->partial[page_nr].offset;
201 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200202 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200203 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200204 if (spd->flags & SPLICE_F_GIFT)
205 buf->flags |= PIPE_BUF_FLAG_GIFT;
206
Jens Axboe6f767b02006-04-11 13:53:56 +0200207 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200208 page_nr++;
209 ret += buf->len;
210
Jens Axboe6f767b02006-04-11 13:53:56 +0200211 if (pipe->inode)
212 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200213
Jens Axboe912d35f2006-04-26 10:59:21 +0200214 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200215 break;
Jens Axboe6f767b02006-04-11 13:53:56 +0200216 if (pipe->nrbufs < PIPE_BUFFERS)
Jens Axboe5274f052006-03-30 15:15:30 +0200217 continue;
218
219 break;
220 }
221
Jens Axboe912d35f2006-04-26 10:59:21 +0200222 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700223 if (!ret)
224 ret = -EAGAIN;
225 break;
226 }
227
Jens Axboe5274f052006-03-30 15:15:30 +0200228 if (signal_pending(current)) {
229 if (!ret)
230 ret = -ERESTARTSYS;
231 break;
232 }
233
234 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200235 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200236 if (waitqueue_active(&pipe->wait))
237 wake_up_interruptible_sync(&pipe->wait);
238 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200239 do_wakeup = 0;
240 }
241
Ingo Molnar3a326a22006-04-10 15:18:35 +0200242 pipe->waiting_writers++;
243 pipe_wait(pipe);
244 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200245 }
246
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200247 pipe_unlock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200248
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200249 if (do_wakeup) {
250 smp_mb();
251 if (waitqueue_active(&pipe->wait))
252 wake_up_interruptible(&pipe->wait);
253 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200254 }
255
Jens Axboe00de00b2007-06-15 13:14:22 +0200256 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800257 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200258
259 return ret;
260}
261
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800262static void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
263{
264 page_cache_release(spd->pages[i]);
265}
266
Ingo Molnar3a326a22006-04-10 15:18:35 +0200267static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200268__generic_file_splice_read(struct file *in, loff_t *ppos,
269 struct pipe_inode_info *pipe, size_t len,
270 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200271{
272 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700273 unsigned int loff, nr_pages, req_pages;
Jens Axboe16c523d2006-04-10 09:03:58 +0200274 struct page *pages[PIPE_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +0200275 struct partial_page partial[PIPE_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200276 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200277 pgoff_t index, end_index;
278 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200279 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200280 struct splice_pipe_desc spd = {
281 .pages = pages,
282 .partial = partial,
283 .flags = flags,
284 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800285 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200286 };
Jens Axboe5274f052006-03-30 15:15:30 +0200287
Jens Axboecbb7e572006-04-11 14:57:50 +0200288 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200289 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700290 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
291 nr_pages = min(req_pages, (unsigned)PIPE_BUFFERS);
Jens Axboe5274f052006-03-30 15:15:30 +0200292
293 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200294 * Lookup the (hopefully) full range of pages we need.
295 */
296 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700297 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200298
299 /*
300 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a4822007-07-19 01:48:05 -0700301 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200302 */
Fengguang Wu431a4822007-07-19 01:48:05 -0700303 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700304 page_cache_sync_readahead(mapping, &in->f_ra, in,
305 index, req_pages - spd.nr_pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700306
Jens Axboe932cc6d2007-06-21 13:10:21 +0200307 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200308 while (spd.nr_pages < nr_pages) {
309 /*
310 * Page could be there, find_get_pages_contig() breaks on
311 * the first hole.
312 */
313 page = find_get_page(mapping, index);
314 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200315 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200316 * page didn't exist, allocate one.
317 */
318 page = page_cache_alloc_cold(mapping);
319 if (!page)
320 break;
321
322 error = add_to_page_cache_lru(page, mapping, index,
Hugh Dickins4cd13502008-04-03 23:35:22 +0100323 mapping_gfp_mask(mapping));
Jens Axboeeb207962006-04-27 11:05:22 +0200324 if (unlikely(error)) {
325 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200326 if (error == -EEXIST)
327 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200328 break;
329 }
330 /*
331 * add_to_page_cache() locks the page, unlock it
332 * to avoid convoluting the logic below even more.
333 */
334 unlock_page(page);
335 }
336
337 pages[spd.nr_pages++] = page;
338 index++;
339 }
340
341 /*
342 * Now loop over the map and see if we need to start IO on any
343 * pages, fill in the partial map, etc.
344 */
345 index = *ppos >> PAGE_CACHE_SHIFT;
346 nr_pages = spd.nr_pages;
347 spd.nr_pages = 0;
348 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200349 unsigned int this_len;
350
351 if (!len)
352 break;
353
354 /*
355 * this_len is the max we'll use from this page
356 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200357 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboeeb207962006-04-27 11:05:22 +0200358 page = pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200359
Fengguang Wua08a1662007-07-19 01:48:03 -0700360 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700361 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700362 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700363
Jens Axboe7480a902006-04-11 13:52:47 +0200364 /*
365 * If the page isn't uptodate, we may need to start io on it
366 */
367 if (!PageUptodate(page)) {
Jens Axboec4f895c2006-04-19 15:56:12 +0200368 /*
369 * If in nonblock mode then dont block on waiting
370 * for an in-flight io page
371 */
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200372 if (flags & SPLICE_F_NONBLOCK) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200373 if (!trylock_page(page)) {
Jens Axboe8191ecd2008-04-10 08:24:25 +0200374 error = -EAGAIN;
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200375 break;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200376 }
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200377 } else
378 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200379
380 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200381 * Page was truncated, or invalidated by the
382 * filesystem. Redo the find/create, but this time the
383 * page is kept locked, so there's no chance of another
384 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200385 */
386 if (!page->mapping) {
387 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200388 page = find_or_create_page(mapping, index,
389 mapping_gfp_mask(mapping));
390
391 if (!page) {
392 error = -ENOMEM;
393 break;
394 }
395 page_cache_release(pages[page_nr]);
396 pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200397 }
398 /*
399 * page was already under io and is now done, great
400 */
401 if (PageUptodate(page)) {
402 unlock_page(page);
403 goto fill_it;
404 }
405
Jens Axboe7480a902006-04-11 13:52:47 +0200406 /*
407 * need to read in the page
408 */
409 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200410 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200411 /*
412 * We really should re-lookup the page here,
413 * but it complicates things a lot. Instead
414 * lets just do what we already stored, and
415 * we'll get it the next time we are called.
416 */
Jens Axboe7480a902006-04-11 13:52:47 +0200417 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200418 error = 0;
419
Jens Axboe7480a902006-04-11 13:52:47 +0200420 break;
421 }
Jens Axboe620a3242007-06-07 09:39:42 +0200422 }
423fill_it:
424 /*
425 * i_size must be checked after PageUptodate.
426 */
427 isize = i_size_read(mapping->host);
428 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
429 if (unlikely(!isize || index > end_index))
430 break;
431
432 /*
433 * if this is the last page, see if we need to shrink
434 * the length and stop
435 */
436 if (end_index == index) {
437 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200438
439 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200440 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200441 */
Jens Axboe620a3242007-06-07 09:39:42 +0200442 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
443 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200444 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200445
446 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200447 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200448 */
Jens Axboe620a3242007-06-07 09:39:42 +0200449 this_len = min(this_len, plen - loff);
450 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200451 }
Jens Axboe620a3242007-06-07 09:39:42 +0200452
Jens Axboeeb207962006-04-27 11:05:22 +0200453 partial[page_nr].offset = loff;
454 partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200455 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200456 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200457 spd.nr_pages++;
458 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200459 }
460
Jens Axboeeb207962006-04-27 11:05:22 +0200461 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200462 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200463 * we got, 'nr_pages' is how many pages are in the map.
464 */
465 while (page_nr < nr_pages)
466 page_cache_release(pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700467 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200468
Jens Axboe912d35f2006-04-26 10:59:21 +0200469 if (spd.nr_pages)
Jens Axboe00522fb2006-04-26 14:39:29 +0200470 return splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200471
Jens Axboe7480a902006-04-11 13:52:47 +0200472 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200473}
474
Jens Axboe83f91352006-04-02 23:05:09 +0200475/**
476 * generic_file_splice_read - splice data from file to a pipe
477 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200478 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200479 * @pipe: pipe to splice to
480 * @len: number of bytes to splice
481 * @flags: splice modifier flags
482 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200483 * Description:
484 * Will read pages from given file and fill them into a pipe. Can be
485 * used as long as the address_space operations for the source implements
486 * a readpage() hook.
487 *
Jens Axboe83f91352006-04-02 23:05:09 +0200488 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200489ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
490 struct pipe_inode_info *pipe, size_t len,
491 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200492{
Jens Axboed366d3982007-06-01 14:54:11 +0200493 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200494 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200495
496 isize = i_size_read(in->f_mapping->host);
497 if (unlikely(*ppos >= isize))
498 return 0;
499
500 left = isize - *ppos;
501 if (unlikely(left < len))
502 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200503
Jens Axboe8191ecd2008-04-10 08:24:25 +0200504 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200505 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200506 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200507 file_accessed(in);
508 }
Jens Axboe5274f052006-03-30 15:15:30 +0200509
510 return ret;
511}
Jens Axboe059a8f32006-04-02 23:06:05 +0200512EXPORT_SYMBOL(generic_file_splice_read);
513
Miklos Szeredi68181732009-05-07 15:37:36 +0200514static const struct pipe_buf_operations default_pipe_buf_ops = {
515 .can_merge = 0,
516 .map = generic_pipe_buf_map,
517 .unmap = generic_pipe_buf_unmap,
518 .confirm = generic_pipe_buf_confirm,
519 .release = generic_pipe_buf_release,
520 .steal = generic_pipe_buf_steal,
521 .get = generic_pipe_buf_get,
522};
523
524static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
525 unsigned long vlen, loff_t offset)
526{
527 mm_segment_t old_fs;
528 loff_t pos = offset;
529 ssize_t res;
530
531 old_fs = get_fs();
532 set_fs(get_ds());
533 /* The cast to a user pointer is valid due to the set_fs() */
534 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
535 set_fs(old_fs);
536
537 return res;
538}
539
Miklos Szeredib2858d72009-05-19 11:37:46 +0200540static ssize_t kernel_write(struct file *file, const char *buf, size_t count,
541 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200542{
543 mm_segment_t old_fs;
544 ssize_t res;
545
546 old_fs = get_fs();
547 set_fs(get_ds());
548 /* The cast to a user pointer is valid due to the set_fs() */
Miklos Szeredib2858d72009-05-19 11:37:46 +0200549 res = vfs_write(file, (const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200550 set_fs(old_fs);
551
552 return res;
553}
554
Miklos Szeredi68181732009-05-07 15:37:36 +0200555ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
556 struct pipe_inode_info *pipe, size_t len,
557 unsigned int flags)
558{
559 unsigned int nr_pages;
560 unsigned int nr_freed;
561 size_t offset;
562 struct page *pages[PIPE_BUFFERS];
563 struct partial_page partial[PIPE_BUFFERS];
564 struct iovec vec[PIPE_BUFFERS];
565 pgoff_t index;
566 ssize_t res;
567 size_t this_len;
568 int error;
569 int i;
570 struct splice_pipe_desc spd = {
571 .pages = pages,
572 .partial = partial,
573 .flags = flags,
574 .ops = &default_pipe_buf_ops,
575 .spd_release = spd_release_page,
576 };
577
578 index = *ppos >> PAGE_CACHE_SHIFT;
579 offset = *ppos & ~PAGE_CACHE_MASK;
580 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
581
582 for (i = 0; i < nr_pages && i < PIPE_BUFFERS && len; i++) {
583 struct page *page;
584
Jens Axboe4f231222009-05-13 08:35:35 +0200585 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200586 error = -ENOMEM;
587 if (!page)
588 goto err;
589
590 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200591 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200592 vec[i].iov_len = this_len;
593 pages[i] = page;
594 spd.nr_pages++;
595 len -= this_len;
596 offset = 0;
597 }
598
599 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200600 if (res < 0) {
601 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200602 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200603 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200604
605 error = 0;
606 if (!res)
607 goto err;
608
609 nr_freed = 0;
610 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200611 this_len = min_t(size_t, vec[i].iov_len, res);
612 partial[i].offset = 0;
613 partial[i].len = this_len;
614 if (!this_len) {
615 __free_page(pages[i]);
616 pages[i] = NULL;
617 nr_freed++;
618 }
619 res -= this_len;
620 }
621 spd.nr_pages -= nr_freed;
622
623 res = splice_to_pipe(pipe, &spd);
624 if (res > 0)
625 *ppos += res;
626
627 return res;
628
629err:
Jens Axboe4f231222009-05-13 08:35:35 +0200630 for (i = 0; i < spd.nr_pages; i++)
Miklos Szeredi68181732009-05-07 15:37:36 +0200631 __free_page(pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200632
Miklos Szeredi68181732009-05-07 15:37:36 +0200633 return error;
634}
635EXPORT_SYMBOL(default_file_splice_read);
636
Jens Axboe5274f052006-03-30 15:15:30 +0200637/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200638 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200639 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200640 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200641static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200642 struct pipe_buffer *buf, struct splice_desc *sd)
643{
Jens Axboe6a14b902007-06-14 13:08:55 +0200644 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200645 loff_t pos = sd->pos;
Jens Axboef84d7512006-05-01 19:59:03 +0200646 int ret, more;
Jens Axboe5274f052006-03-30 15:15:30 +0200647
Jens Axboecac36bb2007-06-14 13:10:48 +0200648 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200649 if (!ret) {
650 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200651
Jens Axboef84d7512006-05-01 19:59:03 +0200652 ret = file->f_op->sendpage(file, buf->page, buf->offset,
653 sd->len, &pos, more);
654 }
Jens Axboe5274f052006-03-30 15:15:30 +0200655
Jens Axboe016b6612006-04-25 15:42:00 +0200656 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200657}
658
659/*
660 * This is a little more tricky than the file -> pipe splicing. There are
661 * basically three cases:
662 *
663 * - Destination page already exists in the address space and there
664 * are users of it. For that case we have no other option that
665 * copying the data. Tough luck.
666 * - Destination page already exists in the address space, but there
667 * are no users of it. Make sure it's uptodate, then drop it. Fall
668 * through to last case.
669 * - Destination page does not exist, we can add the pipe page to
670 * the page cache and avoid the copy.
671 *
Jens Axboe83f91352006-04-02 23:05:09 +0200672 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
673 * sd->flags), we attempt to migrate pages from the pipe to the output
674 * file address space page cache. This is possible if no one else has
675 * the pipe page referenced outside of the pipe and page cache. If
676 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
677 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200678 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200679int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
680 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200681{
Jens Axboe6a14b902007-06-14 13:08:55 +0200682 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200683 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200684 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200685 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700686 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200687 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200688
689 /*
Jens Axboe49d0b212006-04-10 09:04:41 +0200690 * make sure the data in this buffer is uptodate
Jens Axboe5274f052006-03-30 15:15:30 +0200691 */
Jens Axboecac36bb2007-06-14 13:10:48 +0200692 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200693 if (unlikely(ret))
694 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200695
Jens Axboe5274f052006-03-30 15:15:30 +0200696 offset = sd->pos & ~PAGE_CACHE_MASK;
697
Jens Axboe016b6612006-04-25 15:42:00 +0200698 this_len = sd->len;
699 if (this_len + offset > PAGE_CACHE_SIZE)
700 this_len = PAGE_CACHE_SIZE - offset;
701
Nick Pigginafddba42007-10-16 01:25:01 -0700702 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
703 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
704 if (unlikely(ret))
705 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200706
Jens Axboe0568b402006-05-01 19:50:48 +0200707 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200708 /*
709 * Careful, ->map() uses KM_USER0!
710 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200711 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200712 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200713
Jens Axboe016b6612006-04-25 15:42:00 +0200714 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200715 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200716 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200717 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200718 }
Nick Pigginafddba42007-10-16 01:25:01 -0700719 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
720 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200721out:
Jens Axboe5274f052006-03-30 15:15:30 +0200722 return ret;
723}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200724EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200725
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200726static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
727{
728 smp_mb();
729 if (waitqueue_active(&pipe->wait))
730 wake_up_interruptible(&pipe->wait);
731 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
732}
733
734/**
735 * splice_from_pipe_feed - feed available data from a pipe to a file
736 * @pipe: pipe to splice from
737 * @sd: information to @actor
738 * @actor: handler that splices the data
739 *
740 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200741 * This function loops over the pipe and calls @actor to do the
742 * actual moving of a single struct pipe_buffer to the desired
743 * destination. It returns when there's no more buffers left in
744 * the pipe or if the requested number of bytes (@sd->total_len)
745 * have been copied. It returns a positive number (one) if the
746 * pipe needs to be filled with more data, zero if the required
747 * number of bytes have been copied and -errno on error.
748 *
749 * This, together with splice_from_pipe_{begin,end,next}, may be
750 * used to implement the functionality of __splice_from_pipe() when
751 * locking is required around copying the pipe buffers to the
752 * destination.
753 */
754int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
755 splice_actor *actor)
756{
757 int ret;
758
759 while (pipe->nrbufs) {
760 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
761 const struct pipe_buf_operations *ops = buf->ops;
762
763 sd->len = buf->len;
764 if (sd->len > sd->total_len)
765 sd->len = sd->total_len;
766
767 ret = actor(pipe, buf, sd);
768 if (ret <= 0) {
769 if (ret == -ENODATA)
770 ret = 0;
771 return ret;
772 }
773 buf->offset += ret;
774 buf->len -= ret;
775
776 sd->num_spliced += ret;
777 sd->len -= ret;
778 sd->pos += ret;
779 sd->total_len -= ret;
780
781 if (!buf->len) {
782 buf->ops = NULL;
783 ops->release(pipe, buf);
784 pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1);
785 pipe->nrbufs--;
786 if (pipe->inode)
787 sd->need_wakeup = true;
788 }
789
790 if (!sd->total_len)
791 return 0;
792 }
793
794 return 1;
795}
796EXPORT_SYMBOL(splice_from_pipe_feed);
797
798/**
799 * splice_from_pipe_next - wait for some data to splice from
800 * @pipe: pipe to splice from
801 * @sd: information about the splice operation
802 *
803 * Description:
804 * This function will wait for some data and return a positive
805 * value (one) if pipe buffers are available. It will return zero
806 * or -errno if no more data needs to be spliced.
807 */
808int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
809{
810 while (!pipe->nrbufs) {
811 if (!pipe->writers)
812 return 0;
813
814 if (!pipe->waiting_writers && sd->num_spliced)
815 return 0;
816
817 if (sd->flags & SPLICE_F_NONBLOCK)
818 return -EAGAIN;
819
820 if (signal_pending(current))
821 return -ERESTARTSYS;
822
823 if (sd->need_wakeup) {
824 wakeup_pipe_writers(pipe);
825 sd->need_wakeup = false;
826 }
827
828 pipe_wait(pipe);
829 }
830
831 return 1;
832}
833EXPORT_SYMBOL(splice_from_pipe_next);
834
835/**
836 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700837 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200838 *
839 * Description:
840 * This function should be called before a loop containing
841 * splice_from_pipe_next() and splice_from_pipe_feed() to
842 * initialize the necessary fields of @sd.
843 */
844void splice_from_pipe_begin(struct splice_desc *sd)
845{
846 sd->num_spliced = 0;
847 sd->need_wakeup = false;
848}
849EXPORT_SYMBOL(splice_from_pipe_begin);
850
851/**
852 * splice_from_pipe_end - finish splicing from pipe
853 * @pipe: pipe to splice from
854 * @sd: information about the splice operation
855 *
856 * Description:
857 * This function will wake up pipe writers if necessary. It should
858 * be called after a loop containing splice_from_pipe_next() and
859 * splice_from_pipe_feed().
860 */
861void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
862{
863 if (sd->need_wakeup)
864 wakeup_pipe_writers(pipe);
865}
866EXPORT_SYMBOL(splice_from_pipe_end);
867
Jens Axboe932cc6d2007-06-21 13:10:21 +0200868/**
869 * __splice_from_pipe - splice data from a pipe to given actor
870 * @pipe: pipe to splice from
871 * @sd: information to @actor
872 * @actor: handler that splices the data
873 *
874 * Description:
875 * This function does little more than loop over the pipe and call
876 * @actor to do the actual moving of a single struct pipe_buffer to
877 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
878 * pipe_to_user.
879 *
Jens Axboe83f91352006-04-02 23:05:09 +0200880 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200881ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
882 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200883{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200884 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200885
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200886 splice_from_pipe_begin(sd);
887 do {
888 ret = splice_from_pipe_next(pipe, sd);
889 if (ret > 0)
890 ret = splice_from_pipe_feed(pipe, sd, actor);
891 } while (ret > 0);
892 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200893
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200894 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200895}
Mark Fasheh40bee442007-03-21 13:11:02 +0100896EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200897
Jens Axboe932cc6d2007-06-21 13:10:21 +0200898/**
899 * splice_from_pipe - splice data from a pipe to a file
900 * @pipe: pipe to splice from
901 * @out: file to splice to
902 * @ppos: position in @out
903 * @len: how many bytes to splice
904 * @flags: splice modifier flags
905 * @actor: handler that splices the data
906 *
907 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200908 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200909 * otherwise it's identical to __splice_from_pipe().
910 *
911 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200912ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
913 loff_t *ppos, size_t len, unsigned int flags,
914 splice_actor *actor)
915{
916 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200917 struct splice_desc sd = {
918 .total_len = len,
919 .flags = flags,
920 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200921 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200922 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200923
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200924 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200925 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200926 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200927
928 return ret;
929}
930
931/**
Jens Axboe83f91352006-04-02 23:05:09 +0200932 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200933 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200934 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200935 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200936 * @len: number of bytes to splice
937 * @flags: splice modifier flags
938 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200939 * Description:
940 * Will either move or copy pages (determined by @flags options) from
941 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200942 *
943 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200944ssize_t
945generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200946 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200947{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200948 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200949 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200950 struct splice_desc sd = {
951 .total_len = len,
952 .flags = flags,
953 .pos = *ppos,
954 .u.file = out,
955 };
Ingo Molnar3a326a22006-04-10 15:18:35 +0200956 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200957
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200958 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +0200959
960 splice_from_pipe_begin(&sd);
961 do {
962 ret = splice_from_pipe_next(pipe, &sd);
963 if (ret <= 0)
964 break;
965
966 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
967 ret = file_remove_suid(out);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200968 if (!ret) {
969 file_update_time(out);
Miklos Szeredieb443e52009-04-14 19:48:38 +0200970 ret = splice_from_pipe_feed(pipe, &sd, pipe_to_file);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200971 }
Miklos Szeredieb443e52009-04-14 19:48:38 +0200972 mutex_unlock(&inode->i_mutex);
973 } while (ret > 0);
974 splice_from_pipe_end(pipe, &sd);
975
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200976 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +0200977
978 if (sd.num_spliced)
979 ret = sd.num_spliced;
980
Jens Axboea4514eb2006-04-19 15:57:05 +0200981 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200982 unsigned long nr_pages;
Jan Kara148f9482009-08-17 19:52:36 +0200983 int err;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200984
Jens Axboe17ee4f42007-06-15 13:10:37 +0200985 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200986
Jan Kara148f9482009-08-17 19:52:36 +0200987 err = generic_write_sync(out, *ppos, ret);
988 if (err)
989 ret = err;
990 else
991 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200992 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200993 }
994
995 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200996}
997
Jens Axboe059a8f32006-04-02 23:06:05 +0200998EXPORT_SYMBOL(generic_file_splice_write);
999
Miklos Szeredib2858d72009-05-19 11:37:46 +02001000static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1001 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001002{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001003 int ret;
1004 void *data;
1005
1006 ret = buf->ops->confirm(pipe, buf);
1007 if (ret)
1008 return ret;
1009
1010 data = buf->ops->map(pipe, buf, 0);
1011 ret = kernel_write(sd->u.file, data + buf->offset, sd->len, sd->pos);
1012 buf->ops->unmap(pipe, buf, data);
1013
1014 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001015}
1016
1017static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1018 struct file *out, loff_t *ppos,
1019 size_t len, unsigned int flags)
1020{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001021 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001022
Miklos Szeredib2858d72009-05-19 11:37:46 +02001023 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1024 if (ret > 0)
1025 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001026
Miklos Szeredib2858d72009-05-19 11:37:46 +02001027 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001028}
1029
Jens Axboe83f91352006-04-02 23:05:09 +02001030/**
1031 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001032 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001033 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001034 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001035 * @len: number of bytes to splice
1036 * @flags: splice modifier flags
1037 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001038 * Description:
1039 * Will send @len bytes from the pipe to a network socket. No data copying
1040 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001041 *
1042 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001043ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001044 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001045{
Jens Axboe00522fb2006-04-26 14:39:29 +02001046 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001047}
1048
Jens Axboe059a8f32006-04-02 23:06:05 +02001049EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001050
Jens Axboe83f91352006-04-02 23:05:09 +02001051/*
1052 * Attempt to initiate a splice from pipe to file.
1053 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001054static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001055 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001056{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001057 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1058 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001059 int ret;
1060
Jens Axboe49570e92006-04-11 13:56:09 +02001061 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +02001062 return -EBADF;
1063
Linus Torvaldsefc968d2008-10-09 14:04:54 -07001064 if (unlikely(out->f_flags & O_APPEND))
1065 return -EINVAL;
1066
Jens Axboecbb7e572006-04-11 14:57:50 +02001067 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001068 if (unlikely(ret < 0))
1069 return ret;
1070
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001071 splice_write = out->f_op->splice_write;
1072 if (!splice_write)
1073 splice_write = default_file_splice_write;
1074
1075 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001076}
1077
Jens Axboe83f91352006-04-02 23:05:09 +02001078/*
1079 * Attempt to initiate a splice from a file to a pipe.
1080 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001081static long do_splice_to(struct file *in, loff_t *ppos,
1082 struct pipe_inode_info *pipe, size_t len,
1083 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001084{
Miklos Szeredi68181732009-05-07 15:37:36 +02001085 ssize_t (*splice_read)(struct file *, loff_t *,
1086 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001087 int ret;
1088
Jens Axboe49570e92006-04-11 13:56:09 +02001089 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001090 return -EBADF;
1091
Jens Axboecbb7e572006-04-11 14:57:50 +02001092 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001093 if (unlikely(ret < 0))
1094 return ret;
1095
Miklos Szeredi68181732009-05-07 15:37:36 +02001096 splice_read = in->f_op->splice_read;
1097 if (!splice_read)
1098 splice_read = default_file_splice_read;
1099
1100 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001101}
1102
Jens Axboe932cc6d2007-06-21 13:10:21 +02001103/**
1104 * splice_direct_to_actor - splices data directly between two non-pipes
1105 * @in: file to splice from
1106 * @sd: actor information on where to splice to
1107 * @actor: handles the data splicing
1108 *
1109 * Description:
1110 * This is a special case helper to splice directly between two
1111 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001112 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001113 * that process.
1114 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001115 */
1116ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1117 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001118{
1119 struct pipe_inode_info *pipe;
1120 long ret, bytes;
1121 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001122 size_t len;
1123 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001124
1125 /*
1126 * We require the input being a regular file, as we don't want to
1127 * randomly drop data for eg socket -> socket splicing. Use the
1128 * piped splicing for that!
1129 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001130 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001131 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1132 return -EINVAL;
1133
1134 /*
1135 * neither in nor out is a pipe, setup an internal pipe attached to
1136 * 'out' and transfer the wanted data from 'in' to 'out' through that
1137 */
1138 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001139 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001140 pipe = alloc_pipe_info(NULL);
1141 if (!pipe)
1142 return -ENOMEM;
1143
1144 /*
1145 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001146 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001147 * PIPE_READERS appropriately.
1148 */
1149 pipe->readers = 1;
1150
1151 current->splice_pipe = pipe;
1152 }
1153
1154 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001155 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001156 */
1157 ret = 0;
1158 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001159 len = sd->total_len;
1160 flags = sd->flags;
1161
1162 /*
1163 * Don't block on output, we have to drain the direct pipe.
1164 */
1165 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001166
1167 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001168 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001169 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001170
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001171 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001172 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001173 goto out_release;
1174
1175 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001176 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001177
1178 /*
1179 * NOTE: nonblocking mode only applies to the input. We
1180 * must not do the output in nonblocking mode as then we
1181 * could get stuck data in the internal pipe:
1182 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001183 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001184 if (unlikely(ret <= 0)) {
1185 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001186 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001187 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001188
1189 bytes += ret;
1190 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001191 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001192
Tom Zanussia82c53a2008-05-09 13:28:36 +02001193 if (ret < read_len) {
1194 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001195 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001196 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001197 }
1198
Jens Axboe9e971982008-01-29 21:05:57 +01001199done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001200 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001201 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001202 return bytes;
1203
1204out_release:
1205 /*
1206 * If we did an incomplete transfer we must release
1207 * the pipe buffers in question:
1208 */
1209 for (i = 0; i < PIPE_BUFFERS; i++) {
1210 struct pipe_buffer *buf = pipe->bufs + i;
1211
1212 if (buf->ops) {
1213 buf->ops->release(pipe, buf);
1214 buf->ops = NULL;
1215 }
1216 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001217
Jens Axboe9e971982008-01-29 21:05:57 +01001218 if (!bytes)
1219 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001220
Jens Axboe9e971982008-01-29 21:05:57 +01001221 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001222}
1223EXPORT_SYMBOL(splice_direct_to_actor);
1224
1225static int direct_splice_actor(struct pipe_inode_info *pipe,
1226 struct splice_desc *sd)
1227{
Jens Axboe6a14b902007-06-14 13:08:55 +02001228 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001229
1230 return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags);
1231}
1232
Jens Axboe932cc6d2007-06-21 13:10:21 +02001233/**
1234 * do_splice_direct - splices data directly between two files
1235 * @in: file to splice from
1236 * @ppos: input file offset
1237 * @out: file to splice to
1238 * @len: number of bytes to splice
1239 * @flags: splice modifier flags
1240 *
1241 * Description:
1242 * For use by do_sendfile(). splice can easily emulate sendfile, but
1243 * doing it in the application would incur an extra system call
1244 * (splice in + splice out, as compared to just sendfile()). So this helper
1245 * can splice directly through a process-private pipe.
1246 *
1247 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001248long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1249 size_t len, unsigned int flags)
1250{
1251 struct splice_desc sd = {
1252 .len = len,
1253 .total_len = len,
1254 .flags = flags,
1255 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001256 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001257 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001258 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001259
1260 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001261 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001262 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001263
Jens Axboec66ab6f2007-06-12 21:17:17 +02001264 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001265}
1266
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001267static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1268 struct pipe_inode_info *opipe,
1269 size_t len, unsigned int flags);
Jens Axboe83f91352006-04-02 23:05:09 +02001270/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001271 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1272 * location, so checking ->i_pipe is not enough to verify that this is a
1273 * pipe.
1274 */
1275static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1276{
1277 if (S_ISFIFO(inode->i_mode))
1278 return inode->i_pipe;
1279
1280 return NULL;
1281}
1282
1283/*
Jens Axboe83f91352006-04-02 23:05:09 +02001284 * Determine where to splice to/from.
1285 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001286static long do_splice(struct file *in, loff_t __user *off_in,
1287 struct file *out, loff_t __user *off_out,
1288 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001289{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001290 struct pipe_inode_info *ipipe;
1291 struct pipe_inode_info *opipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001292 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001293 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001294
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001295 ipipe = pipe_info(in->f_path.dentry->d_inode);
1296 opipe = pipe_info(out->f_path.dentry->d_inode);
1297
1298 if (ipipe && opipe) {
1299 if (off_in || off_out)
1300 return -ESPIPE;
1301
1302 if (!(in->f_mode & FMODE_READ))
1303 return -EBADF;
1304
1305 if (!(out->f_mode & FMODE_WRITE))
1306 return -EBADF;
1307
1308 /* Splicing to self would be fun, but... */
1309 if (ipipe == opipe)
1310 return -EINVAL;
1311
1312 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1313 }
1314
1315 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001316 if (off_in)
1317 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001318 if (off_out) {
1319 if (out->f_op->llseek == no_llseek)
1320 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001321 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001322 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001323 off = &offset;
1324 } else
1325 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001326
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001327 ret = do_splice_from(ipipe, out, off, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001328
1329 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1330 ret = -EFAULT;
1331
1332 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001333 }
Jens Axboe5274f052006-03-30 15:15:30 +02001334
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001335 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001336 if (off_out)
1337 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001338 if (off_in) {
1339 if (in->f_op->llseek == no_llseek)
1340 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001341 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001342 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001343 off = &offset;
1344 } else
1345 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001346
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001347 ret = do_splice_to(in, off, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001348
1349 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1350 ret = -EFAULT;
1351
1352 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001353 }
Jens Axboe5274f052006-03-30 15:15:30 +02001354
1355 return -EINVAL;
1356}
1357
Jens Axboe912d35f2006-04-26 10:59:21 +02001358/*
1359 * Map an iov into an array of pages and offset/length tupples. With the
1360 * partial_page structure, we can map several non-contiguous ranges into
1361 * our ones pages[] map instead of splitting that operation into pieces.
1362 * Could easily be exported as a generic helper for other users, in which
1363 * case one would probably want to add a 'max_nr_pages' parameter as well.
1364 */
1365static int get_iovec_page_array(const struct iovec __user *iov,
1366 unsigned int nr_vecs, struct page **pages,
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001367 struct partial_page *partial, int aligned)
Jens Axboe912d35f2006-04-26 10:59:21 +02001368{
1369 int buffers = 0, error = 0;
1370
Jens Axboe912d35f2006-04-26 10:59:21 +02001371 while (nr_vecs) {
1372 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001373 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001374 void __user *base;
1375 size_t len;
1376 int i;
1377
Linus Torvalds75723952007-10-01 13:17:28 -07001378 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001379 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001380 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001381
1382 base = entry.iov_base;
1383 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001384
1385 /*
1386 * Sanity check this iovec. 0 read succeeds.
1387 */
Linus Torvalds75723952007-10-01 13:17:28 -07001388 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001389 if (unlikely(!len))
1390 break;
1391 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001392 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001393 break;
1394
1395 /*
1396 * Get this base offset and number of pages, then map
1397 * in the user pages.
1398 */
1399 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001400
1401 /*
1402 * If asked for alignment, the offset must be zero and the
1403 * length a multiple of the PAGE_SIZE.
1404 */
1405 error = -EINVAL;
1406 if (aligned && (off || len & ~PAGE_MASK))
1407 break;
1408
Jens Axboe912d35f2006-04-26 10:59:21 +02001409 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1410 if (npages > PIPE_BUFFERS - buffers)
1411 npages = PIPE_BUFFERS - buffers;
1412
Nick Pigginbc40d732008-07-25 19:45:26 -07001413 error = get_user_pages_fast((unsigned long)base, npages,
1414 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001415
1416 if (unlikely(error <= 0))
1417 break;
1418
1419 /*
1420 * Fill this contiguous range into the partial page map.
1421 */
1422 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001423 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001424
1425 partial[buffers].offset = off;
1426 partial[buffers].len = plen;
1427
1428 off = 0;
1429 len -= plen;
1430 buffers++;
1431 }
1432
1433 /*
1434 * We didn't complete this iov, stop here since it probably
1435 * means we have to move some of this into a pipe to
1436 * be able to continue.
1437 */
1438 if (len)
1439 break;
1440
1441 /*
1442 * Don't continue if we mapped fewer pages than we asked for,
1443 * or if we mapped the max number of pages that we have
1444 * room for.
1445 */
1446 if (error < npages || buffers == PIPE_BUFFERS)
1447 break;
1448
1449 nr_vecs--;
1450 iov++;
1451 }
1452
Jens Axboe912d35f2006-04-26 10:59:21 +02001453 if (buffers)
1454 return buffers;
1455
1456 return error;
1457}
1458
Jens Axboe6a14b902007-06-14 13:08:55 +02001459static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1460 struct splice_desc *sd)
1461{
1462 char *src;
1463 int ret;
1464
Jens Axboecac36bb2007-06-14 13:10:48 +02001465 ret = buf->ops->confirm(pipe, buf);
Jens Axboe6a14b902007-06-14 13:08:55 +02001466 if (unlikely(ret))
1467 return ret;
1468
1469 /*
1470 * See if we can use the atomic maps, by prefaulting in the
1471 * pages and doing an atomic copy
1472 */
1473 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1474 src = buf->ops->map(pipe, buf, 1);
1475 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1476 sd->len);
1477 buf->ops->unmap(pipe, buf, src);
1478 if (!ret) {
1479 ret = sd->len;
1480 goto out;
1481 }
1482 }
1483
1484 /*
1485 * No dice, use slow non-atomic map and copy
1486 */
1487 src = buf->ops->map(pipe, buf, 0);
1488
1489 ret = sd->len;
1490 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1491 ret = -EFAULT;
1492
Jens Axboe6866bef2007-10-16 10:01:29 +02001493 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001494out:
1495 if (ret > 0)
1496 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001497 return ret;
1498}
1499
1500/*
1501 * For lack of a better implementation, implement vmsplice() to userspace
1502 * as a simple copy of the pipes pages to the user iov.
1503 */
1504static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1505 unsigned long nr_segs, unsigned int flags)
1506{
1507 struct pipe_inode_info *pipe;
1508 struct splice_desc sd;
1509 ssize_t size;
1510 int error;
1511 long ret;
1512
1513 pipe = pipe_info(file->f_path.dentry->d_inode);
1514 if (!pipe)
1515 return -EBADF;
1516
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001517 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001518
1519 error = ret = 0;
1520 while (nr_segs) {
1521 void __user *base;
1522 size_t len;
1523
1524 /*
1525 * Get user address base and length for this iovec.
1526 */
1527 error = get_user(base, &iov->iov_base);
1528 if (unlikely(error))
1529 break;
1530 error = get_user(len, &iov->iov_len);
1531 if (unlikely(error))
1532 break;
1533
1534 /*
1535 * Sanity check this iovec. 0 read succeeds.
1536 */
1537 if (unlikely(!len))
1538 break;
1539 if (unlikely(!base)) {
1540 error = -EFAULT;
1541 break;
1542 }
1543
Jens Axboe88119302008-02-08 08:49:14 -08001544 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1545 error = -EFAULT;
1546 break;
1547 }
1548
Jens Axboe6a14b902007-06-14 13:08:55 +02001549 sd.len = 0;
1550 sd.total_len = len;
1551 sd.flags = flags;
1552 sd.u.userptr = base;
1553 sd.pos = 0;
1554
1555 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1556 if (size < 0) {
1557 if (!ret)
1558 ret = size;
1559
1560 break;
1561 }
1562
1563 ret += size;
1564
1565 if (size < len)
1566 break;
1567
1568 nr_segs--;
1569 iov++;
1570 }
1571
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001572 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001573
1574 if (!ret)
1575 ret = error;
1576
1577 return ret;
1578}
1579
Jens Axboe912d35f2006-04-26 10:59:21 +02001580/*
1581 * vmsplice splices a user address range into a pipe. It can be thought of
1582 * as splice-from-memory, where the regular splice is splice-from-file (or
1583 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001584 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001585static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1586 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001587{
Jens Axboeddac0d32006-11-04 12:49:32 +01001588 struct pipe_inode_info *pipe;
Jens Axboe912d35f2006-04-26 10:59:21 +02001589 struct page *pages[PIPE_BUFFERS];
1590 struct partial_page partial[PIPE_BUFFERS];
1591 struct splice_pipe_desc spd = {
1592 .pages = pages,
1593 .partial = partial,
1594 .flags = flags,
1595 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001596 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001597 };
1598
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001599 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001600 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001601 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001602
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001603 spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial,
1604 flags & SPLICE_F_GIFT);
Jens Axboe912d35f2006-04-26 10:59:21 +02001605 if (spd.nr_pages <= 0)
1606 return spd.nr_pages;
1607
Jens Axboe00522fb2006-04-26 14:39:29 +02001608 return splice_to_pipe(pipe, &spd);
Jens Axboe912d35f2006-04-26 10:59:21 +02001609}
1610
Jens Axboe6a14b902007-06-14 13:08:55 +02001611/*
1612 * Note that vmsplice only really supports true splicing _from_ user memory
1613 * to a pipe, not the other way around. Splicing from user memory is a simple
1614 * operation that can be supported without any funky alignment restrictions
1615 * or nasty vm tricks. We simply map in the user memory and fill them into
1616 * a pipe. The reverse isn't quite as easy, though. There are two possible
1617 * solutions for that:
1618 *
1619 * - memcpy() the data internally, at which point we might as well just
1620 * do a regular read() on the buffer anyway.
1621 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1622 * has restriction limitations on both ends of the pipe).
1623 *
1624 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1625 *
1626 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001627SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1628 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001629{
1630 struct file *file;
1631 long error;
1632 int fput;
1633
Jens Axboe6a14b902007-06-14 13:08:55 +02001634 if (unlikely(nr_segs > UIO_MAXIOV))
1635 return -EINVAL;
1636 else if (unlikely(!nr_segs))
1637 return 0;
1638
Jens Axboe912d35f2006-04-26 10:59:21 +02001639 error = -EBADF;
1640 file = fget_light(fd, &fput);
1641 if (file) {
1642 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001643 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1644 else if (file->f_mode & FMODE_READ)
1645 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001646
1647 fput_light(file, fput);
1648 }
1649
1650 return error;
1651}
1652
Heiko Carstens836f92a2009-01-14 14:14:33 +01001653SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1654 int, fd_out, loff_t __user *, off_out,
1655 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001656{
1657 long error;
1658 struct file *in, *out;
1659 int fput_in, fput_out;
1660
1661 if (unlikely(!len))
1662 return 0;
1663
1664 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001665 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001666 if (in) {
1667 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001668 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001669 if (out) {
1670 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001671 error = do_splice(in, off_in,
1672 out, off_out,
1673 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001674 fput_light(out, fput_out);
1675 }
1676 }
1677
1678 fput_light(in, fput_in);
1679 }
1680
1681 return error;
1682}
Jens Axboe70524492006-04-11 15:51:17 +02001683
1684/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001685 * Make sure there's data to read. Wait for input if we can, otherwise
1686 * return an appropriate error.
1687 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001688static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001689{
1690 int ret;
1691
1692 /*
1693 * Check ->nrbufs without the inode lock first. This function
1694 * is speculative anyways, so missing one is ok.
1695 */
1696 if (pipe->nrbufs)
1697 return 0;
1698
1699 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001700 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001701
1702 while (!pipe->nrbufs) {
1703 if (signal_pending(current)) {
1704 ret = -ERESTARTSYS;
1705 break;
1706 }
1707 if (!pipe->writers)
1708 break;
1709 if (!pipe->waiting_writers) {
1710 if (flags & SPLICE_F_NONBLOCK) {
1711 ret = -EAGAIN;
1712 break;
1713 }
1714 }
1715 pipe_wait(pipe);
1716 }
1717
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001718 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001719 return ret;
1720}
1721
1722/*
1723 * Make sure there's writeable room. Wait for room if we can, otherwise
1724 * return an appropriate error.
1725 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001726static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001727{
1728 int ret;
1729
1730 /*
1731 * Check ->nrbufs without the inode lock first. This function
1732 * is speculative anyways, so missing one is ok.
1733 */
1734 if (pipe->nrbufs < PIPE_BUFFERS)
1735 return 0;
1736
1737 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001738 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001739
1740 while (pipe->nrbufs >= PIPE_BUFFERS) {
1741 if (!pipe->readers) {
1742 send_sig(SIGPIPE, current, 0);
1743 ret = -EPIPE;
1744 break;
1745 }
1746 if (flags & SPLICE_F_NONBLOCK) {
1747 ret = -EAGAIN;
1748 break;
1749 }
1750 if (signal_pending(current)) {
1751 ret = -ERESTARTSYS;
1752 break;
1753 }
1754 pipe->waiting_writers++;
1755 pipe_wait(pipe);
1756 pipe->waiting_writers--;
1757 }
1758
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001759 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001760 return ret;
1761}
1762
1763/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001764 * Splice contents of ipipe to opipe.
1765 */
1766static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1767 struct pipe_inode_info *opipe,
1768 size_t len, unsigned int flags)
1769{
1770 struct pipe_buffer *ibuf, *obuf;
1771 int ret = 0, nbuf;
1772 bool input_wakeup = false;
1773
1774
1775retry:
1776 ret = ipipe_prep(ipipe, flags);
1777 if (ret)
1778 return ret;
1779
1780 ret = opipe_prep(opipe, flags);
1781 if (ret)
1782 return ret;
1783
1784 /*
1785 * Potential ABBA deadlock, work around it by ordering lock
1786 * grabbing by pipe info address. Otherwise two different processes
1787 * could deadlock (one doing tee from A -> B, the other from B -> A).
1788 */
1789 pipe_double_lock(ipipe, opipe);
1790
1791 do {
1792 if (!opipe->readers) {
1793 send_sig(SIGPIPE, current, 0);
1794 if (!ret)
1795 ret = -EPIPE;
1796 break;
1797 }
1798
1799 if (!ipipe->nrbufs && !ipipe->writers)
1800 break;
1801
1802 /*
1803 * Cannot make any progress, because either the input
1804 * pipe is empty or the output pipe is full.
1805 */
1806 if (!ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS) {
1807 /* Already processed some buffers, break */
1808 if (ret)
1809 break;
1810
1811 if (flags & SPLICE_F_NONBLOCK) {
1812 ret = -EAGAIN;
1813 break;
1814 }
1815
1816 /*
1817 * We raced with another reader/writer and haven't
1818 * managed to process any buffers. A zero return
1819 * value means EOF, so retry instead.
1820 */
1821 pipe_unlock(ipipe);
1822 pipe_unlock(opipe);
1823 goto retry;
1824 }
1825
1826 ibuf = ipipe->bufs + ipipe->curbuf;
1827 nbuf = (opipe->curbuf + opipe->nrbufs) % PIPE_BUFFERS;
1828 obuf = opipe->bufs + nbuf;
1829
1830 if (len >= ibuf->len) {
1831 /*
1832 * Simply move the whole buffer from ipipe to opipe
1833 */
1834 *obuf = *ibuf;
1835 ibuf->ops = NULL;
1836 opipe->nrbufs++;
1837 ipipe->curbuf = (ipipe->curbuf + 1) % PIPE_BUFFERS;
1838 ipipe->nrbufs--;
1839 input_wakeup = true;
1840 } else {
1841 /*
1842 * Get a reference to this pipe buffer,
1843 * so we can copy the contents over.
1844 */
1845 ibuf->ops->get(ipipe, ibuf);
1846 *obuf = *ibuf;
1847
1848 /*
1849 * Don't inherit the gift flag, we need to
1850 * prevent multiple steals of this page.
1851 */
1852 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1853
1854 obuf->len = len;
1855 opipe->nrbufs++;
1856 ibuf->offset += obuf->len;
1857 ibuf->len -= obuf->len;
1858 }
1859 ret += obuf->len;
1860 len -= obuf->len;
1861 } while (len);
1862
1863 pipe_unlock(ipipe);
1864 pipe_unlock(opipe);
1865
1866 /*
1867 * If we put data in the output pipe, wakeup any potential readers.
1868 */
1869 if (ret > 0) {
1870 smp_mb();
1871 if (waitqueue_active(&opipe->wait))
1872 wake_up_interruptible(&opipe->wait);
1873 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1874 }
1875 if (input_wakeup)
1876 wakeup_pipe_writers(ipipe);
1877
1878 return ret;
1879}
1880
1881/*
Jens Axboe70524492006-04-11 15:51:17 +02001882 * Link contents of ipipe to opipe.
1883 */
1884static int link_pipe(struct pipe_inode_info *ipipe,
1885 struct pipe_inode_info *opipe,
1886 size_t len, unsigned int flags)
1887{
1888 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001889 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001890
1891 /*
1892 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001893 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001894 * could deadlock (one doing tee from A -> B, the other from B -> A).
1895 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001896 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001897
Jens Axboeaadd06e2006-07-10 11:00:01 +02001898 do {
Jens Axboe70524492006-04-11 15:51:17 +02001899 if (!opipe->readers) {
1900 send_sig(SIGPIPE, current, 0);
1901 if (!ret)
1902 ret = -EPIPE;
1903 break;
1904 }
Jens Axboe70524492006-04-11 15:51:17 +02001905
1906 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001907 * If we have iterated all input buffers or ran out of
1908 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001909 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001910 if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS)
Jens Axboe70524492006-04-11 15:51:17 +02001911 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001912
1913 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1));
1914 nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1);
1915
Jens Axboe2a272502006-04-19 15:56:40 +02001916 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001917 * Get a reference to this pipe buffer,
1918 * so we can copy the contents over.
Jens Axboe2a272502006-04-19 15:56:40 +02001919 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001920 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001921
Jens Axboeaadd06e2006-07-10 11:00:01 +02001922 obuf = opipe->bufs + nbuf;
1923 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001924
Jens Axboeaadd06e2006-07-10 11:00:01 +02001925 /*
1926 * Don't inherit the gift flag, we need to
1927 * prevent multiple steals of this page.
1928 */
1929 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1930
1931 if (obuf->len > len)
1932 obuf->len = len;
1933
1934 opipe->nrbufs++;
1935 ret += obuf->len;
1936 len -= obuf->len;
1937 i++;
1938 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001939
Jens Axboe02cf01a2008-02-20 10:34:51 +01001940 /*
1941 * return EAGAIN if we have the potential of some data in the
1942 * future, otherwise just return 0
1943 */
1944 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1945 ret = -EAGAIN;
1946
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001947 pipe_unlock(ipipe);
1948 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001949
Jens Axboeaadd06e2006-07-10 11:00:01 +02001950 /*
1951 * If we put data in the output pipe, wakeup any potential readers.
1952 */
1953 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02001954 smp_mb();
1955 if (waitqueue_active(&opipe->wait))
1956 wake_up_interruptible(&opipe->wait);
1957 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1958 }
1959
1960 return ret;
1961}
1962
1963/*
1964 * This is a tee(1) implementation that works on pipes. It doesn't copy
1965 * any data, it simply references the 'in' pages on the 'out' pipe.
1966 * The 'flags' used are the SPLICE_F_* variants, currently the only
1967 * applicable one is SPLICE_F_NONBLOCK.
1968 */
1969static long do_tee(struct file *in, struct file *out, size_t len,
1970 unsigned int flags)
1971{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001972 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
1973 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001974 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001975
1976 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001977 * Duplicate the contents of ipipe to opipe without actually
1978 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001979 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001980 if (ipipe && opipe && ipipe != opipe) {
1981 /*
1982 * Keep going, unless we encounter an error. The ipipe/opipe
1983 * ordering doesn't really matter.
1984 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001985 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001986 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001987 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01001988 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001989 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001990 }
1991 }
Jens Axboe70524492006-04-11 15:51:17 +02001992
Jens Axboeaadd06e2006-07-10 11:00:01 +02001993 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02001994}
1995
Heiko Carstens836f92a2009-01-14 14:14:33 +01001996SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02001997{
1998 struct file *in;
1999 int error, fput_in;
2000
2001 if (unlikely(!len))
2002 return 0;
2003
2004 error = -EBADF;
2005 in = fget_light(fdin, &fput_in);
2006 if (in) {
2007 if (in->f_mode & FMODE_READ) {
2008 int fput_out;
2009 struct file *out = fget_light(fdout, &fput_out);
2010
2011 if (out) {
2012 if (out->f_mode & FMODE_WRITE)
2013 error = do_tee(in, out, len, flags);
2014 fput_light(out, fput_out);
2015 }
2016 }
2017 fput_light(in, fput_in);
2018 }
2019
2020 return error;
2021}